Bernstein: Sandisk (SNDK - Outperform, $300 TP) Sandisk sees the CES keynote comments on Vera Rubin KV cache from Jensen Huang of Nvidia as incremental new NAND demand that is not previously in the demand forecast. Prior to the forecast, Sandisk estimated that NAND was 8% undersupplied and this incremental demand will just make the extremely tight situation even tighter. The company couldn't quantify the impact as it was quite new information but felt that the impact was significant. Sandisk forecasted Datacenter NAND demand growth at 45% vs. mid 20s% for overall NAND bit demand growth compared to supply growth in the teens%. We estimate the KV cache could indicate an additional 16TB per GPU, vs. 3-4TB per GPU today. This would equate to >5x jump in NAND content. The company reaffirmed that they are not adding capacity and doesn't see any significant new capacity coming online besides some from YMTC. Sandisk is currently taking a cautious approach towards supply additions following a few years of low profits. They need to see demand consistently outstrip supply with margins significantly higher than this before even considering capacity additions. In addition Sandisk is careful to consider the long-term impact of supply additions, which are essentially permanent. There was some discussion around competitors and how YMTC is really the only NAND player that is adding capacity and isn't acting rationally (arguably). Long term agreements are becoming a hot topic but currently less than 10% of business is locked in on longer-term price deals. Therefore, over 90% of SNDK's business is on short term (i.e. one quarter or less) price agreements. In an up-market this is a good thing as they are therefore more levered to the market price increases. Sandisk mentioned that they are rising prices faster than the company ever has. Prices are increasing substantially every week in a manner that is unprecedented in the memory industry. Where NAND supply bit growth is coming from: • For SNDK its entirely from technology migration to BiCS 8, which is offset by loss in wafer throughput due to the added layers. This in turn is offset by expansion in clean room space. • NAND industry supply growth is in mid to high teens, which is SNDK's goal. For 2026, SNDK may be below that. • Very few cleanrooms available for NAND expansion besides SNDK/Kioxia's Kitakami Fab (but this is planned to offset wafer throughput loss from BiCS8 ramp), and YMTC. Any new cleanrooms would take 2-3 years before having significant impact on the market due to the long lead times. Where all the NAND demand growth is coming from: • Disproportionally most of the incremental demand growth is coming from datacenters, which in turn is coming from AI. Datacenters will be the largest portion of overall NAND demand for the first time this year. • Traditional NAND markets like mobile, have more price sensitivity than datacenter customers, which don't appear to be sensitive to price increases. • Although the rapid price increase is not expected to impact datacenter demand, overall NAND demand will be impacted due to some demand impact in mobile and PC markets (in the form of lower content growth or lower unit growth or both. Even with this weakness the market is still expected to be very tight due to the strong and price insensitive datacenter customers. High Bandwidth Flash (HBF) showing promise and remaining on track: • HBF is a flash based replacement for HBM used for inference. ...